swap.c

Linux 1.0 内核C源代码 Linux最早版本代码 由Linus Torvalds亲自书写的

		swap_task = 1;
		goto check_task;
	}
	p = task[swap_task];
	if (!p || !p->swappable) {
		swap_task++;
		goto check_task;
	}
check_dir:
	if (swap_table >= PTRS_PER_PAGE) {
		swap_table = 0;
		swap_task++;
		goto check_task;
	}
	pg_table = ((unsigned long *) p->tss.cr3)[swap_table];
	if (pg_table >= high_memory || (mem_map[MAP_NR(pg_table)] & MAP_PAGE_RESERVED)) {
		swap_table++;
		goto check_dir;
	}
	if (!(PAGE_PRESENT & pg_table)) {
		printk("bad page-table at pg_dir[%d]: %08x\n",
			swap_table,pg_table);
		((unsigned long *) p->tss.cr3)[swap_table] = 0;
		swap_table++;
		goto check_dir;
	}
	pg_table &= PAGE_MASK;
check_table:
	if (swap_page >= PTRS_PER_PAGE) {
		swap_page = 0;
		swap_table++;
		goto check_dir;
	}
	switch (try_to_swap_out(swap_page + (unsigned long *) pg_table)) {
		case 0: break;
		case 1: p->rss--; return 1;
		default: p->rss--;
	}
	swap_page++;
	goto check_table;
}

#endif

static int try_to_free_page(void)
{
	int i=6;

	while (i--) {
		if (shrink_buffers(i))
			return 1;
		if (shm_swap(i))
			return 1;
		if (swap_out(i))
			return 1;
	}
	return 0;
}

/*
 * Note that this must be atomic, or bad things will happen when
 * pages are requested in interrupts (as malloc can do). Thus the
 * cli/sti's.
 */
static inline void add_mem_queue(unsigned long addr, unsigned long * queue)
{
	addr &= PAGE_MASK;
	*(unsigned long *) addr = *queue;
	*queue = addr;
}

/*
 * Free_page() adds the page to the free lists. This is optimized for
 * fast normal cases (no error jumps taken normally).
 *
 * The way to optimize jumps for gcc-2.2.2 is to:
 *  - select the "normal" case and put it inside the if () { XXX }
 *  - no else-statements if you can avoid them
 *
 * With the above two rules, you get a straight-line execution path
 * for the normal case, giving better asm-code.
 */
void free_page(unsigned long addr)
{
	if (addr < high_memory) {
		unsigned short * map = mem_map + MAP_NR(addr);

		if (*map) {
			if (!(*map & MAP_PAGE_RESERVED)) {
				unsigned long flag;

				save_flags(flag);
				cli();
				if (!--*map) {
					if (nr_secondary_pages < MAX_SECONDARY_PAGES) {
						add_mem_queue(addr,&secondary_page_list);
						nr_secondary_pages++;
						restore_flags(flag);
						return;
					}
					add_mem_queue(addr,&free_page_list);
					nr_free_pages++;
				}
				restore_flags(flag);
			}
			return;
		}
		printk("Trying to free free memory (%08lx): memory probabably corrupted\n",addr);
		printk("PC = %08lx\n",*(((unsigned long *)&addr)-1));
		return;
	}
}

/*
 * This is one ugly macro, but it simplifies checking, and makes
 * this speed-critical place reasonably fast, especially as we have
 * to do things with the interrupt flag etc.
 *
 * Note that this #define is heavily optimized to give fast code
 * for the normal case - the if-statements are ordered so that gcc-2.2.2
 * will make *no* jumps for the normal code. Don't touch unless you
 * know what you are doing.
 */
#define REMOVE_FROM_MEM_QUEUE(queue,nr) \
	cli(); \
	if ((result = queue) != 0) { \
		if (!(result & ~PAGE_MASK) && result < high_memory) { \
			queue = *(unsigned long *) result; \
			if (!mem_map[MAP_NR(result)]) { \
				mem_map[MAP_NR(result)] = 1; \
				nr--; \
last_free_pages[index = (index + 1) & (NR_LAST_FREE_PAGES - 1)] = result; \
				restore_flags(flag); \
				return result; \
			} \
			printk("Free page %08lx has mem_map = %d\n", \
				result,mem_map[MAP_NR(result)]); \
		} else \
			printk("Result = 0x%08lx - memory map destroyed\n", result); \
		queue = 0; \
		nr = 0; \
	} else if (nr) { \
		printk(#nr " is %d, but " #queue " is empty\n",nr); \
		nr = 0; \
	} \
	restore_flags(flag)

/*
 * Get physical address of first (actually last :-) free page, and mark it
 * used. If no free pages left, return 0.
 *
 * Note that this is one of the most heavily called functions in the kernel,
 * so it's a bit timing-critical (especially as we have to disable interrupts
 * in it). See the above macro which does most of the work, and which is
 * optimized for a fast normal path of execution.
 */
unsigned long __get_free_page(int priority)
{
	extern unsigned long intr_count;
	unsigned long result, flag;
	static unsigned long index = 0;

	/* this routine can be called at interrupt time via
	   malloc.  We want to make sure that the critical
	   sections of code have interrupts disabled. -RAB
	   Is this code reentrant? */

	if (intr_count && priority != GFP_ATOMIC) {
		printk("gfp called nonatomically from interrupt %08lx\n",
			((unsigned long *)&priority)[-1]);
		priority = GFP_ATOMIC;
	}
	save_flags(flag);
repeat:
	REMOVE_FROM_MEM_QUEUE(free_page_list,nr_free_pages);
	if (priority == GFP_BUFFER)
		return 0;
	if (priority != GFP_ATOMIC)
		if (try_to_free_page())
			goto repeat;
	REMOVE_FROM_MEM_QUEUE(secondary_page_list,nr_secondary_pages);
	return 0;
}

/*
 * Trying to stop swapping from a file is fraught with races, so
 * we repeat quite a bit here when we have to pause. swapoff()
 * isn't exactly timing-critical, so who cares?
 */
static int try_to_unuse(unsigned int type)
{
	int nr, pgt, pg;
	unsigned long page, *ppage;
	unsigned long tmp = 0;
	struct task_struct *p;

	nr = 0;
/*
 * When we have to sleep, we restart the whole algorithm from the same
 * task we stopped in. That at least rids us of all races.
 */
repeat:
	for (; nr < NR_TASKS ; nr++) {
		p = task[nr];
		if (!p)
			continue;
		for (pgt = 0 ; pgt < PTRS_PER_PAGE ; pgt++) {
			ppage = pgt + ((unsigned long *) p->tss.cr3);
			page = *ppage;
			if (!page)
				continue;
			if (!(page & PAGE_PRESENT) || (page >= high_memory))
				continue;
			if (mem_map[MAP_NR(page)] & MAP_PAGE_RESERVED)
				continue;
			ppage = (unsigned long *) (page & PAGE_MASK);	
			for (pg = 0 ; pg < PTRS_PER_PAGE ; pg++,ppage++) {
				page = *ppage;
				if (!page)
					continue;
				if (page & PAGE_PRESENT)
					continue;
				if (SWP_TYPE(page) != type)
					continue;
				if (!tmp) {
					if (!(tmp = __get_free_page(GFP_KERNEL)))
						return -ENOMEM;
					goto repeat;
				}
				read_swap_page(page, (char *) tmp);
				if (*ppage == page) {
					*ppage = tmp | (PAGE_DIRTY | PAGE_PRIVATE);
					++p->rss;
					swap_free(page);
					tmp = 0;
				}
				goto repeat;
			}
		}
	}
	free_page(tmp);
	return 0;
}

asmlinkage int sys_swapoff(const char * specialfile)
{
	struct swap_info_struct * p;
	struct inode * inode;
	unsigned int type;
	int i;

	if (!suser())
		return -EPERM;
	i = namei(specialfile,&inode);
	if (i)
		return i;
	p = swap_info;
	for (type = 0 ; type < nr_swapfiles ; type++,p++) {
		if ((p->flags & SWP_WRITEOK) != SWP_WRITEOK)
			continue;
		if (p->swap_file) {
			if (p->swap_file == inode)
				break;
		} else {
			if (!S_ISBLK(inode->i_mode))
				continue;
			if (p->swap_device == inode->i_rdev)
				break;
		}
	}
	iput(inode);
	if (type >= nr_swapfiles)
		return -EINVAL;
	p->flags = SWP_USED;
	i = try_to_unuse(type);
	if (i) {
		p->flags = SWP_WRITEOK;
		return i;
	}
	nr_swap_pages -= p->pages;
	iput(p->swap_file);
	p->swap_file = NULL;
	p->swap_device = 0;
	vfree(p->swap_map);
	p->swap_map = NULL;
	free_page((long) p->swap_lockmap);
	p->swap_lockmap = NULL;
	p->flags = 0;
	return 0;
}

/*
 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
 *
 * The swapon system call
 */
asmlinkage int sys_swapon(const char * specialfile)
{
	struct swap_info_struct * p;
	struct inode * swap_inode;
	unsigned int type;
	int i,j;
	int error;

	if (!suser())
		return -EPERM;
	p = swap_info;
	for (type = 0 ; type < nr_swapfiles ; type++,p++)
		if (!(p->flags & SWP_USED))
			break;
	if (type >= MAX_SWAPFILES)
		return -EPERM;
	if (type >= nr_swapfiles)
		nr_swapfiles = type+1;
	p->flags = SWP_USED;
	p->swap_file = NULL;
	p->swap_device = 0;
	p->swap_map = NULL;
	p->swap_lockmap = NULL;
	p->lowest_bit = 0;
	p->highest_bit = 0;
	p->max = 1;
	error = namei(specialfile,&swap_inode);
	if (error)
		goto bad_swap;
	error = -EBUSY;
	if (swap_inode->i_count != 1)
		goto bad_swap;
	error = -EINVAL;
	if (S_ISBLK(swap_inode->i_mode)) {
		p->swap_device = swap_inode->i_rdev;
		iput(swap_inode);
		error = -ENODEV;
		if (!p->swap_device)
			goto bad_swap;
		error = -EBUSY;
		for (i = 0 ; i < nr_swapfiles ; i++) {
			if (i == type)
				continue;
			if (p->swap_device == swap_info[i].swap_device)
				goto bad_swap;
		}
	} else if (S_ISREG(swap_inode->i_mode))
		p->swap_file = swap_inode;
	else
		goto bad_swap;
	p->swap_lockmap = (unsigned char *) get_free_page(GFP_USER);
	if (!p->swap_lockmap) {
		printk("Unable to start swapping: out of memory :-)\n");
		error = -ENOMEM;
		goto bad_swap;
	}
	read_swap_page(SWP_ENTRY(type,0), (char *) p->swap_lockmap);
	if (memcmp("SWAP-SPACE",p->swap_lockmap+4086,10)) {
		printk("Unable to find swap-space signature\n");
		error = -EINVAL;
		goto bad_swap;
	}
	memset(p->swap_lockmap+PAGE_SIZE-10,0,10);
	j = 0;
	p->lowest_bit = 0;
	p->highest_bit = 0;
	for (i = 1 ; i < 8*PAGE_SIZE ; i++) {
		if (test_bit(i,p->swap_lockmap)) {
			if (!p->lowest_bit)
				p->lowest_bit = i;
			p->highest_bit = i;
			p->max = i+1;
			j++;
		}
	}
	if (!j) {
		printk("Empty swap-file\n");
		error = -EINVAL;
		goto bad_swap;
	}
	p->swap_map = (unsigned char *) vmalloc(p->max);
	if (!p->swap_map) {
		error = -ENOMEM;
		goto bad_swap;
	}
	for (i = 1 ; i < p->max ; i++) {
		if (test_bit(i,p->swap_lockmap))
			p->swap_map[i] = 0;
		else
			p->swap_map[i] = 0x80;
	}
	p->swap_map[0] = 0x80;
	memset(p->swap_lockmap,0,PAGE_SIZE);
	p->flags = SWP_WRITEOK;
	p->pages = j;
	nr_swap_pages += j;
	printk("Adding Swap: %dk swap-space\n",j<<2);
	return 0;
bad_swap:
	free_page((long) p->swap_lockmap);
	vfree(p->swap_map);
	iput(p->swap_file);
	p->swap_device = 0;
	p->swap_file = NULL;
	p->swap_map = NULL;
	p->swap_lockmap = NULL;
	p->flags = 0;
	return error;
}

void si_swapinfo(struct sysinfo *val)
{
	unsigned int i, j;

	val->freeswap = val->totalswap = 0;
	for (i = 0; i < nr_swapfiles; i++) {
		if (!(swap_info[i].flags & SWP_USED))
			continue;
		for (j = 0; j < swap_info[i].max; ++j)
			switch (swap_info[i].swap_map[j]) {
				case 128:
					continue;
				case 0:
					++val->freeswap;
				default:
					++val->totalswap;
			}
	}
	val->freeswap <<= PAGE_SHIFT;
	val->totalswap <<= PAGE_SHIFT;
	return;
}